The Web contains a huge amount of information in its unstructured texts. Analyzing these texts is very important as more and more people post product reviews at merchant sites, discussion groups, etc. This paper presents a set of language patterns, which is composed of 22 rules, to extract two-noun phrases from customer reviews. Two-noun phrases are specific and interesting when compared with one-noun words. Normally, these phrases contain product features which are useful for customers. Three tagging methods are used to generate part- of-speech tags for Bing Liu's dataset. On average, the recall of each tagging method is above 90 percent no matter what the tagging method is. With this set of rules in hand, we can keep 22 percent or more two-noun phrases from being extracted in each category, which are useless and do not need to be extracted. Additionally, language rules can be used to extract some useful product features that human taggers fail to annotate.